topic Re: Estimated means in SAS Procedures

Estimated means

Iron_grief — Sat, 04 Sep 2010 14:36:20 GMT

I am quiet new to SAS and I apologize in advance if it is a silly question. I am trying to obtain estimated proportions for a categorical variable, but it seems that I can only get confidence interval if a specify a grouping variable. Suppose I have catVAR1 whose proportion is 0.20, I want to adjust the proportion for age and gender. I have to put gender into the CLASS statement and in the LSMEANS statement of the GLM procedure, like this:

PROC GLM;
CLASS gender;
MODEL catVAR1 = gender age;
LSMEANS gender / CL;
RUN;

This way i get the proportions of catVAR1 for males and females. What if I want the overall proportion in the population, only adjusted for age and gender? if I remove gender from the LSMEANS statement, I get the adjusted proportion but not the CI.
Thanks in advance.

Iron.

Re: Estimated means

Doc_Duke — Sun, 05 Sep 2010 17:33:24 GMT

You are using the wrong statistical procedure; GLM assumes normal distribution on the residuals. You need to be using PROC LOGISTIC.

Doc Muhlbaier
Duke

Re: Estimated means

Iron_grief — Mon, 06 Sep 2010 15:05:10 GMT

Thanks for your reply, but what about this?

http://www2.sas.com/proceedings/sugi31/140-31.pdf

Example 6 page 5, they use SURVEYREG to obtain adjusted proportion.
Also, does the SURVEYLOGISTIC procedure have a ESTIMATE statement?

Best regards. Message was edited by: Iron_grief

Re: Estimated means

Doc_Duke — Tue, 07 Sep 2010 14:54:25 GMT

I've never used SURVEYREG. I did look at the example and saw that the sample size was nearly 8000. When your sample size is that large, the Central Limit Theorem allows you to use a linear regression procedure when a categorical regression procedure might be more appropriate.